[PATCH] powerpc: Add oprofile calltrace support

Add oprofile calltrace support to powerpc. Disable spinlock backtracing
now we can use calltrace info.

(Updated to work on both 32bit and 64bit by me).

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>

5 files changed, 132 insertions, 4 deletions

diff --git a/arch/powerpc/oprofile/Makefile b/arch/powerpc/oprofile/Makefile
index 554cd7c75321..f5f9859a8338 100644
--- a/arch/powerpc/oprofile/Makefile
+++ b/arch/powerpc/oprofile/Makefile
@@ -6,7 +6,7 @@ DRIVER_OBJS := $(addprefix ../../../drivers/oprofile/, \
 		oprofilefs.o oprofile_stats.o \
 		timer_int.o )
 
-oprofile-y := $(DRIVER_OBJS) common.o
+oprofile-y := $(DRIVER_OBJS) common.o backtrace.o
 oprofile-$(CONFIG_PPC64) += op_model_rs64.o op_model_power4.o
 oprofile-$(CONFIG_FSL_BOOKE) += op_model_fsl_booke.o
 oprofile-$(CONFIG_PPC32) += op_model_7450.o
diff --git a/arch/powerpc/oprofile/backtrace.c b/arch/powerpc/oprofile/backtrace.c
new file mode 100644
index 000000000000..75f57bc96b40
--- /dev/null
+++ b/arch/powerpc/oprofile/backtrace.c
@@ -0,0 +1,126 @@
+/**
+ * Copyright (C) 2005 Brian Rogan <bcr6@cornell.edu>, IBM
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+**/
+
+#include <linux/oprofile.h>
+#include <linux/sched.h>
+#include <asm/processor.h>
+#include <asm/uaccess.h>
+
+#define STACK_SP(STACK)		*(STACK)
+
+#define STACK_LR64(STACK)	*((unsigned long *)(STACK) + 2)
+#define STACK_LR32(STACK)	*((unsigned int *)(STACK) + 1)
+
+#ifdef CONFIG_PPC64
+#define STACK_LR(STACK)		STACK_LR64(STACK)
+#else
+#define STACK_LR(STACK)		STACK_LR32(STACK)
+#endif
+
+static unsigned int user_getsp32(unsigned int sp, int is_first)
+{
+	unsigned int stack_frame[2];
+
+	if (!access_ok(VERIFY_READ, sp, sizeof(stack_frame)))
+		return 0;
+
+	/*
+	 * The most likely reason for this is that we returned -EFAULT,
+	 * which means that we've done all that we can do from
+	 * interrupt context.
+	 */
+	if (__copy_from_user_inatomic(stack_frame, (void *)(long)sp,
+					sizeof(stack_frame)))
+		return 0;
+
+	if (!is_first)
+		oprofile_add_trace(STACK_LR32(stack_frame));
+
+	/*
+	 * We do not enforce increasing stack addresses here because
+	 * we may transition to a different stack, eg a signal handler.
+	 */
+	return STACK_SP(stack_frame);
+}
+
+#ifdef CONFIG_PPC64
+static unsigned long user_getsp64(unsigned long sp, int is_first)
+{
+	unsigned long stack_frame[3];
+
+	if (!access_ok(VERIFY_READ, sp, sizeof(stack_frame)))
+		return 0;
+
+	if (__copy_from_user_inatomic(stack_frame, (void *)sp,
+					sizeof(stack_frame)))
+		return 0;
+
+	if (!is_first)
+		oprofile_add_trace(STACK_LR64(stack_frame));
+
+	return STACK_SP(stack_frame);
+}
+#endif
+
+static unsigned long kernel_getsp(unsigned long sp, int is_first)
+{
+	unsigned long *stack_frame = (unsigned long *)sp;
+
+	if (!validate_sp(sp, current, STACK_FRAME_OVERHEAD))
+		return 0;
+
+	if (!is_first)
+		oprofile_add_trace(STACK_LR(stack_frame));
+
+	/*
+	 * We do not enforce increasing stack addresses here because
+	 * we might be transitioning from an interrupt stack to a kernel
+	 * stack. validate_sp() is designed to understand this, so just
+	 * use it.
+	 */
+	return STACK_SP(stack_frame);
+}
+
+void op_powerpc_backtrace(struct pt_regs * const regs, unsigned int depth)
+{
+	unsigned long sp = regs->gpr[1];
+	int first_frame = 1;
+
+	/* We ditch the top stackframe so need to loop through an extra time */
+	depth += 1;
+
+	if (!user_mode(regs)) {
+		while (depth--) {
+			sp = kernel_getsp(sp, first_frame);
+			if (!sp)
+				break;
+			first_frame = 0;
+		}
+	} else {
+#ifdef CONFIG_PPC64
+		if (!test_thread_flag(TIF_32BIT)) {
+			while (depth--) {
+				sp = user_getsp64(sp, first_frame);
+				if (!sp)
+					break;
+				first_frame = 0;
+			}
+
+			return;
+		}
+#endif
+
+		while (depth--) {
+			sp = user_getsp32(sp, first_frame);
+			if (!sp)
+				break;
+			first_frame = 0;
+		}
+	}
+}
diff --git a/arch/powerpc/oprofile/common.c b/arch/powerpc/oprofile/common.c
index cc2535be3a73..2b9143b0f6ba 100644
--- a/arch/powerpc/oprofile/common.c
+++ b/arch/powerpc/oprofile/common.c
@@ -126,8 +126,7 @@ static int op_powerpc_create_files(struct super_block *sb, struct dentry *root)
 	sys.enable_kernel = 1;
 	sys.enable_user = 1;
 #ifdef CONFIG_PPC64
-	/* Turn on backtracing through spinlocks by default */
-	sys.backtrace_spinlocks = 1;
+	sys.backtrace_spinlocks = 0;
 #endif
 
 	return 0;
@@ -168,6 +167,7 @@ int __init oprofile_arch_init(struct oprofile_operations *ops)
 	ops->shutdown = op_powerpc_shutdown;
 	ops->start = op_powerpc_start;
 	ops->stop = op_powerpc_stop;
+	ops->backtrace = op_powerpc_backtrace;
 
 	printk(KERN_INFO "oprofile: using %s performance monitoring.\n",
 	       ops->cpu_type);
diff --git a/arch/powerpc/oprofile/op_model_power4.c b/arch/powerpc/oprofile/op_model_power4.c
index 4b06e53eb9b4..38db2efef3bc 100644
--- a/arch/powerpc/oprofile/op_model_power4.c
+++ b/arch/powerpc/oprofile/op_model_power4.c
@@ -293,7 +293,7 @@ static void power4_handle_interrupt(struct pt_regs *regs,
 		val = ctr_read(i);
 		if (val < 0) {
 			if (oprofile_running && ctr[i].enabled) {
-				oprofile_add_pc(pc, is_kernel, i);
+				oprofile_add_ext_sample(pc, regs, i, is_kernel);
 				ctr_write(i, reset_value[i]);
 			} else {
 				ctr_write(i, 0);
diff --git a/include/asm-powerpc/oprofile_impl.h b/include/asm-powerpc/oprofile_impl.h
index df4defc6321c..aa180e907a67 100644
--- a/include/asm-powerpc/oprofile_impl.h
+++ b/include/asm-powerpc/oprofile_impl.h
@@ -126,5 +126,7 @@ static inline void ctr_write(unsigned int i, unsigned int val)
 }
 #endif /* !CONFIG_FSL_BOOKE */
 
+extern void op_powerpc_backtrace(struct pt_regs * const regs, unsigned int depth);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_POWERPC_OPROFILE_IMPL_H */